Impact of hydrometeor control variables in a convective‐scale 3DEnVar data assimilation scheme

Initialization of condensate water variables for convective‐scale weather forecasting has been a topic of active research for the last few years. In this article, we present the addition of hydrometeor fields as control variables of a three‐dimensional ensemble variational scheme (3DEnVar) for the cloud‐resolving model AROME‐France. Even without any direct assimilation of hydrometeor observations, analysis increments of hydrometeors can be produced via covariances with observed variables in the ensemble‐derived background‐error covariance matrix. Cycled forecast–analysis experiments in near‐operational conditions have been performed over a three‐month summer period. Three configurations are compared: (i) a control experiment without hydrometeor control variables, (ii) a test experiment with hydrometeor control variables but without cycling the resulting forecasts, and (iii) an experiment with hydrometeor variables and cycled forecasts. Compared with the control experiment, both hydrometeor experiments show a positive impact up to 9 hr of forecasts in terms of cloud cover, and in the first hour for precipitation. The spin‐up period, as evidenced by a reduction of precipitation forecast skill at the beginning of the forecasts, is reduced in length and intensity in the hydrometeor experiments. Hydrometeor initialization helps greatly in reducing the spin‐up period usually observed after each assimilation step. This positive impact on spin‐up time is evidenced on this graph by the improvements of precipitation scores at the very beginning of the forecast.